I saw this Kickstarter elsewhere, and I'm frankly amazed they've gotten as much funding as they have.
There aren't that many embarrassingly parallel problems that aren't I/O limited that need to be performed on a consistent basis. If I needed a cluster, then I'd go rent one from Amazon for an hour or four.
For me, the opportunity is really about learning how to develop software for a multicore future, rather than using these for processing. There may not be too many embarrassingly parallel problems around, but that doesn't mean we should avoid developing algorithms for highly scalable and even tightly coupled problems.
I pitched in. There are a tonne of problems that are highly parallel in nature, but need a general purpose processor such as this. You can't just "run everything on a GPU instead" like everyone else is suggesting. Things like if statements are horrendous on GPUs. This gives you the power to scale across, without being limited to GPU architectures. It's also significantly cheaper in the long term than cloud hosting and uses significantly less power than clustering desktop PCs (which is also more expensive, and takes up more space).
3 comments
[ 4.8 ms ] story [ 22.0 ms ] threadThere aren't that many embarrassingly parallel problems that aren't I/O limited that need to be performed on a consistent basis. If I needed a cluster, then I'd go rent one from Amazon for an hour or four.